Auto ML For Adaptive Model Selection And Hyperparameter Optimization

Abstract

Automated Machine Learning (AutoML) improves predictive systems by reducing manual intervention in model configuration and parameter tuning. Existing frameworks often face unstable convergence and limited adaptability in complex hyperparameter search spaces. This research develops an intelligent AutoML framework for adaptive hyperparameter optimization (HPO) and model selection, employing Convolutional Neural Networks (CNN) and Graph Neural Networks (GNN) with a Hybrid Whale-Aquila Optimization (HWAO) algorithm. In HWAO, Whale Optimization performs global exploration to maintain diversity, while Aquila Optimization enhances local exploitation and accelerates convergence toward optimal parameters. This hybrid approach balances exploration and exploitation, reduces premature convergence, and strengthens adaptive learning. Performance evaluation compares standalone Whale and Aquila optimization with the hybrid method. CNN-HWAO achieved an error rate of 11.0%, convergence in 30 iterations, a model size of 115 MB, HPO score of 92.1, and prediction accuracy of 89.0%. GNN-HWAO reached an error rate of 10.2%, convergence in 32 iterations, model size of 125 MB, HPO score of 93.4, and prediction accuracy of 90.1%. The system is trained using Keras 2.2.4 and evaluated with TensorFlow on Python 3.9, demonstrating superior accuracy, efficiency, and generalization.